Event

VLA-Fabric: Multi-arm VLA improves coordination success rates through cross-policy interaction

1 reports · 1 sources

Chenglong Zhang et al. propose VLA-Fabric, which equips each robotic arm with a complete vision-language-action (VLA) model, preserving each arm's local perception and action ownership, and achieves coordination through explicitly learned interaction across policy boundaries. In simulation and real-world experiments, the method was evaluated with three-arm and four-arm teams; on all four real-world tasks, full interaction improved success rates by 16 to 40 percentage points compared with independently trained local policies. The paper systematically studies where interaction should occur, how peer information should be aggregated, and how interaction should be learned, finding that shared context, peer-specific latent variables, and action-stage interaction play complementary coordination roles, and reproduces this principle on VLA architectures with different structures.